Hardening of Acception Links Through Vectorized Lexical Functions

In the framework of the Papillon project, we have defined strategies for populating a pivot dictionnary of interlingual links from monolingual vectorial bases. There are quite a number of acception per entry thus, the proper identification may be quite troublesome and some added clues beside acception links may be usefull. We improve the integrity of the acception base through welsl known semantic relations like synonymy, antonymy, hyperonymy and holonymy relying on lexical functions agents. These semantic relation agents can compute the pertinence of a semantic relation between two acceptions thanks to various lexical informations and conceptual vectors. When a given pertinence score is above a threshold they create a semantic link which can be walked throught by other agents in charge of WSD ot lexical transfert. Base integrity agents walk throw the acceptions and according to their speciality creates semantics links, look for incoherences in the base and emit warning toward lexicographs when neened.